The influence of non-stationary ENSO teleconnections on reconstructions of paleoclimate using a pseudoproxy framework
Abstract. Reconstructions of the El Niño-Southern Oscillation (ENSO) ideally require high-quality, annually-resolved and long-running paleoclimate proxy records in the eastern tropical Pacific Ocean, located in ENSO's centre-of-action. However, to date, the paleoclimate records that have been extracted in the region are short or temporally and spatially sporadic, limiting the information that can be provided by these reconstructions. Consequently, most ENSO reconstructions exploit the downstream influences of ENSO on remote locations, known as teleconnections, where longer records from paleoclimate proxies exist. However, using teleconnections to reconstruct ENSO relies on the assumption that the relationship between ENSO and the remote location is stationary in time. Increasing evidence from observations and climate models suggests that some teleconnections are, in fact, non-stationary, potentially threatening the validity of those paleoclimate reconstructions that exploit teleconnections. This study examines the implications of non-stationary teleconnections on modern multi-proxy reconstructions of ENSO. The sensitivity of the reconstructions to non-stationary teleconnections were tested using a suite of idealized pseudoproxy experiments that employed output from a fully coupled global climate model. Reconstructions of the variance in the Niño 3.4 index, representing ENSO variability, were generated using four different methods to which surface temperature data from the GFDL CM2.1 was applied as a pseudoproxy. As well as sensitivity of the reconstruction to the method, the experiments tested the sensitivity of the reconstruction to the number of non-stationary pseudoproxies and the location of these proxies. ENSO reconstructions in the pseudoproxy experiments were not sensitive to non-stationary teleconnections when global, uniformly-spaced networks of a minimum of approximately 20 proxies were employed. Neglecting proxies from ENSO's center-of-action still produced skillful reconstructions, but the chance of generating a skillful reconstruction decreased. Reconstruction methods that utilized raw time series were the most sensitive to non-stationary teleconnections, while calculating the running variance of pseudoproxies first, appeared to improve the robustness of the resulting reconstructions. The results suggest that caution should be taken when developing reconstructions using proxies from a single teleconnected region, or those that use less than 20 source proxies.